[postgis-devel] PSC Vote: Change PostGIS library name to drop the minor

Wed Aug 30 14:18:05 PDT 2017

Okay now I think I remember what the issue was and Darafei, so I think it has to be a newer or same PostGIS compiled against the same PostgreSQL version (minor for <=9.6, major for >=10)

I think Sandro or Mark may have mentioned this already, but I think what pg_upgrade does is it tries to  recreate all the functions from your old version, but it moves the data as is.  I have to recheck this but it did seem to create a whole bunch of custom scripts that it then deleted when it was done.

So I suspect  it's when it does the 

CREATE FUNCTION …

That it fails if your library was compiled with a lower PostgreSQL or a lower PostGIS than what you have installed.

So as long as we make sure all functions referenced in SQL function in older are still present in the newer lib (more or less ABI compat)

We are okay to keep the same lib name.

Which I think I can say is true from 2.1 -> 2.4

I'm less confident about 2.1 since it's been a while since I've had that. And I think we broke ABI between 2.0 and 2.1.

It would be nice if pg_upgrade had some sort of switch to say – use CREATE EXTENSION instead for things that are in packages (similar to what pg_dump does)  so when it builds the skeleton database to then mount the tables, it doesn't try to reuse the functions that are marked as part of an extension.

That I think should be a switch though because like what Steve said – once you do that there is no going back easily.  I don't know I can think of ways to go back.

I think that would also solve the issue Sandro and I were discussing that a PostGIS 2.4 coming from PostgreSQL 9.4 and upgraded to PostgreSQL 10 is not the same.

It's going to be missing all the parallel features we introduced for 9.6 as well as the rich KNN support we introduced for 9.5.

Thanks,

Regina

From: Regina Obe [mailto:lr at pcorp.us] 
Sent: Wednesday, August 30, 2017 4:52 PM
To: 'PostGIS Development Discussion' <postgis-devel at lists.osgeo.org>
Subject: RE: [postgis-devel] PSC Vote: Change PostGIS library name to drop the minor

Well I'm pretty sure any old library will not do.  I've tried that.

pg_upgrade I think at least checks to see if it can load the main hook function using the new PostgreSQL.  That's why copying the old 2.3 .so from a PostgreSQL 9.6 instance to 10.0 doesn't work.

However I haven't checked to see if making a dummy extension would pass pg_upgrade's test.

So perhaps you can create a postgresql extension called  -- dummy with nothing in it.

And then have  postgis-2.3  be a symlink to dummy.

I suspect that won't work though.  Because I do recall a time I think it was between 2.0 and 2.1 when we made a function change and I tried to do my usual hack of copying the new 2.1 and calling it 2.0 so I could pg_upgrade

And it hiccupped when it couldn't load the function.  That function however might have been a key one that gets called as part of our load process though.  I can't remember what it was exactly.

From: postgis-devel [mailto:postgis-devel-bounces at lists.osgeo.org] On Behalf Of Darafei "Kom?pa" Praliaskouski
Sent: Wednesday, August 30, 2017 5:22 AM
To: PostGIS Development Discussion <postgis-devel at lists.osgeo.org <mailto:postgis-devel at lists.osgeo.org> >
Subject: Re: [postgis-devel] PSC Vote: Change PostGIS library name to drop the minor

I think we need someone to set up experiment, if it is possible to symlink just any library before update, or if it should really match. If it's not called in upgrade and stuff like autovacuum/autoanalyze/whatever uses postgis functions silently are shut down, it may happen that just presence of a shim is enough.

Looking at linux .../lib directories, it's common for liblibrary-1.2.3 to have a symlink from liblibrary-1.2 and from liblibrary-1 and from liblibrary.so. It may also be a way to deal with it.

ср, 30 авг. 2017 г. в 11:48, Sandro Santilli <strk at kbt.io <mailto:strk at kbt.io> >:

On Wed, Aug 30, 2017 at 01:19:03AM -0400, Stephen Frost wrote:

> That is, if you have 9.4 + 2.2 and 9.5 + 2.3, pg_upgrade will refuse to
> run because it's looking in the 9.5 version for the 2.2 .so file and not
> finding it.  Users have to first upgrade to 9.4 + 2.3 and then run
> pg_upgrade to get to 9.5 + 2.3.

But pg_upgrade makes NO ATTEMPT at every running an ALTER EXTENSION,
correct ? That means that *even* if you have a 2.2 .so file under 9.5
you might still be left with half-broken PostGIS (ie: scripts not
using 9.5 only features).

That's why I'm suggesting that pg_upgrade could improve its checking
(since it does implement part of it).
Or allow dropping the initial check (as I think you can still run
ALTER EXTENSION UPDATE while missing the old .so file, right?)

--strk;
_______________________________________________
postgis-devel mailing list
postgis-devel at lists.osgeo.org <mailto:postgis-devel at lists.osgeo.org> 
https://lists.osgeo.org/mailman/listinfo/postgis-devel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-devel/attachments/20170830/abfa0f16/attachment.html>